I
routinely use 3 power amplifiers on the home HiFi. The default amp is a Hugh
Dean AKSA 55W Kit Amplifier with the Nivarna Plus upgrade, This amplifier is a
55W into 8 Ohm (90W into 4 Ohm) Solid State amplifier which has large gobs of
that Valve Amplifier sound and is one of the very few Solid State Amps which I
don’t find to be Cold, Sterile and Boring. For some of my old Rock and Roll
and blues recordings it is however a little bit too revealing and for
listening to these recordings I prefer my little EL84 Ultra linear Push Pull
Valve Amp when I need musical “pace” or my 845 Single Ended Triode Amp
when I just want pure valve warmth and emotion. For some time I have felt the
need for something else. I wanted the pace and excitement of the EL84 amp but
with a bigger (more grunt) sound.
Graham
Maynard’s reference to KT88 Class AB1 ultralinears in his “Class A
Imagineering:Part 1” (Electronics World June 2004) prompted me to get off my
derriere and design and build a prototype using low cost commercial grade
components. I wanted to see what level of performance could be achieved
without spending huge amounts of hard earned cash for toroidal output
transformers and the like. As a further goal I wanted something that any
average DIY’er could build.
You can’t tackle a design without at least a minimum requirements specification. No need to go overboard BUT you need some idea of your goals. Here they are:
Always start a valve power amplifier design from the output and work back to the input.
A search
for readily available (worldwide) output transformers of reasonable cost meant
a Hammond Transformer. The requirements
from above lead me to select the Hammond 1650T.
This transformer is a 1900 Ohms Raa unit with Ultra linear taps at 40%, rated
at 120 Watts and with a FULL POWER bandwidth of 30Hz to 30kHz. The low Raa
(anode to anode primary load impedance) means that 4 off EL34, 6550 or KT88s
can easily drive it to 120 Watts.
I happen
to like EL34s so I started the design with some cheap Chinese EL34s and
immediately ran into a problem. The Ultra linear connection imposes full anode
voltage on the screens of the output tubes. The EL34s suffered screen to
cathode flashovers despite using large series resistors in the screen circuits
to limit screen dissipation. In line with the requirement for robust design I
abandoned EL34s, looked at 6550s and decided their screen rating was also
inadequate and went straight to KT88s (Va = 800V and Vg2 = 600V).
Allowing
for an output transformer efficiency of 90% means we need to deliver about 135
Watts into the primary. That means 506 VRMS or 716V pk across the primary. In
a push pull amp half of this is handled by each side or about 360V pk. KT88s
have a saturation voltage of 90V and so we need a High Voltage rail of 450V as
a minimum. Its actually a good idea to allow 10% more than this so we end up
with a 500V rail.
The
problem with many higher power valve amps is inadequate control of Grid 1.
Most designs tend to use high values of Grid 1 to the bias supply resistors
since these resistors are the “load” for the driver stage. The maximum
recommended value for KT88s in fixed bias is 120K. I used 100K. Each side has
2 of these in parallel (50K) which is a significant load for the phase
splitter/driver stage to handle. Of possibly even more significance is the
total grid capacitance that has to be driven. This value is difficult to
calculate as it consists of grid 1 capacitance plus miller capacitance to the
anode plus miller capacitance to the screen. The gain values required to
calculate it are usually not given on the valve data sheets so I simply
assumed it would be large (approaching 100pF) and proceeded from there.
I already
had a phase splitter design in mind. I wanted to use a current sourced
differential amp. Kevin O’Connor in his “Principals of Power” book
showed a typical circuit using the 2 halves of 12AX7 as a differential
amplifier. Resistors from the two anodes are connected back to the base of the
current source transistor. This arrangement guarantees AC balance as the
current source is “adjusted” at audio frequencies until balance is
achieved. DC balance relies on the two halves of the 12AX7 being well matched.
The 12AX7
is NOT a good choice of signal input device however due to high Miller
Capacitance limiting the input bandwidth. This idea was modified to use a
Cascode connected 6DJ8s (ECC88) for each side of the differential amp similar
to the “traditional” Hedge Circuit. This is where we had a big win. While
doing some net searches I came across Curcio Audio Engineering (CAE) who offer
a PCB with exactly this design combined with a High Voltage Regulator. It is
his PCB1A that was designed for use in upgrading of Dynaco MK3 Valve Amps.
For the MK3 upgrade the diff amp is used to drive the output valve
grids directly. The anodes of the top of the diff amp sit at 200 to 210V DC
which made them ideal for a DC coupled Cathode Followers to be added to drive
the output valve grids, thus there are some coupling caps and bias components
on the PCB1A which you simply don’t load for this application.
I had some
commercial concerns about publishing a design using CAE’s circuit and PCBs
until further net searches showed that the design has been used widely in the
past (eg Sonic Frontiers Amps to name just one) and that the circuit is in the
“public domain”.
That is
enough of a design to build the prototype. With that built we have an ideal
test platform for the final set to work. Rule 1 of any feedback system
(whether it be an audio amp or a laser scanner or whatever) is to make it as
linear as possible before closing the loop.
Note that
I have included a HT fuse and the 10R 0.25W cathode resistors for each output
valve act as individual fuses (as well as giving a convenient point to measure
bias currents). Do NOT be tempted to use high power resistors here.
I
set the output valve bias currents to 50mA per valve and then did plots of
output valve anode voltage and phase vs frequency with a resistive load on the
secondary of the output transformer (and constant grid 1 voltage). This showed
a serious peak at 68 KHz on the “Push” side and an even more complex
arrangement of dips and peaks on the “Pull” side. These resonances need to
be damped by addition of Zobel Networks. I made some enquiries among the valve
amp “gurus” I knew and they all told me that this is a trial and error
procedure and there is no correct way of doing it. For a professional engineer
that was “A Red Rag to a Bull”. Back to the Internet and the reference
books. There has been quite a bit of work done in recent years on critically
damping transformer resonances, mostly in relation to DC-to-DC Converter
design. Some of this work has been interpreted and applied to transformers for
valve amps. The clearest explanation/method I found was on a Valve Amp DIY’ers
site. Here is the link. http://www.siteswithstyle.com/VoltSecond/Damping_ringing_XFMRS/Damping_ringing_in_xfmrs.html.
I used the methods described for maximally flat input impedance (what
VoltSecond calls Rd-opt-Zi) to arrive at the position (anode to screen) and
values of the Zobel Networks shown on the schematic. The voltage and phase
plots where repeated to confirm that critical damping of any ringing had been
achieved. You will note that the capacitor and resistor values are a quite a
bit different than seen on most old published amplifier designs and that they
are different for each side of the transformer. It is IMPORTANT to do this
BEFORE closing the loop as it has a serious impact on the amount of high
frequency roll off required to stabilise the amp once feedback is applied.
Having
done the above, I applied feedback from the 8-Ohm tap of the output
transformer. I found that very minimal step networks across the diff amp anode
resistors were required to stabilise the amp. I then did full power tests
showing 122 Watts at saturation and 138 Watts in heavy overdrive. Output
Impedance without Feedback was 3.5 Ohms and with feedback was 1.8 Ohms. As my
speakers are actually nominally 6 Ohm this was good enough for the Damping
Factor I wanted. Next I did frequency response runs at 50 W output (None of
that frequency response at 1 W rubbish here). At the High Frequency end the
–3dB point was 65kHz which is outstanding. At the low end, response was flat
down to 22 Hz with a severe waveform distortion setting in at 18 Hz due (I
believe) to Output Transformer saturation. Because of this I dropped the
output grid coupling capacitors from 470nF to 100nF to give a –3dB point at
16 Hz (and save some cash). A lag compensation cap across the feedback
resistor is not required however I found that a small value (22pF) did
slightly improve the 10kHz square wave response. I also found that increasing
the KT88s bias currents from 50mA to 55mA per valve made a small but
worthwhile improvement. Increasing it again to 60mA did’nt seem to make any
difference so I settled on 55mA.
While I’m
listening in mono while building the second mono block and so I can’t
actually speak about imaging and the like, the results are simply stunning. It
is by far my favourite amp with a big valve sound. It has the detail of the
845 Single Ended Triode with the pace and excitement of the EL84 UL PP and the
sound is HUGE.
As an
exercise in seeing if a high quality sound, big valve amplifier could be
designed without recourse to fantastically expensive components like toroidal
output transformers, this project has been a huge success. I believe that the
critical point in the process was in optimising the Zobel Networks to
critically damp any ringing in the output transformer. This optimisation
linearised the input impedance and phase of the output transformer allowing it
to perform way past its standard capability. The use of the current source
biased diff amp to guarantee AC balance and the direct coupled cathode
followers to drive the output valve grid capacitance and low value grid
resistances also helped.
You may wish to try doubling
the value of the main high voltage power supply capacitors from 220uF to
470uF. This will add cost BUT may tighten up the bass response a little. The
coupling caps to the output valve grids should be high quality polypropylene
BUT cheap polyesters with a parallel polypropylene cap of about 1/100th
the value should also work well. You might also wish to experiment with the
output valve bias currents. The KT88s should handle idle currents up to 70mA
per valve BUT valve life may be reduced.
Select the Amplifier and Power Supply Schematics
